Build a Social Intelligence Agent with AIsa

The problem

Marketing, product, and community teams want to know how a launch, topic, or competitor is being discussed, and which creators are worth working with. Raw feeds are noisy, and a model left alone will happily invent sentiment, demographics, or trends. A good outcome is an evidence-linked brief: scope and time window, recurring themes with representative links, interpretation labeled as such, and the sampling limits stated. Posting or engaging remains a separate, approved step.

Prompt

Give this to your agent once AIsa is connected via MCP:

Use AIsa to analyze public discussion of <topic or product> on X, Reddit, and YouTube over the last <14> days, in <languages>.
Find the right AIsa tools with `search` and quote paid calls with `batch_quote`. Ask me before any paid call, and tell me how many pages of results you plan to fetch.
For every record keep: platform, canonical URL, author or channel handle, publication time, and the raw metrics returned.
Return: (1) scope and queries used; (2) 3–5 recurring themes, each with 2–3 representative links; (3) recurring questions or complaints; (4) your interpretation, labeled separately; (5) gaps and sampling limits.
Do not infer demographics, sensitive traits, or intent the sources do not state. Do not post, like, follow, or reply.

Variant for creator discovery:

Find 10 creators on YouTube or TikTok who cover <topic>, similar to @example_handle. Show handle, platform, profile URL, and why they fit. Leave out contact emails and AI-inferred demographic fields until I approve the shortlist.

APIs and Skills used

Through MCP, your agent reaches these via a few meta tools at https://mcp.aisa.one/mcp: search, list_categories, get_details, and batch_quote are free; batch_use (up to 20 calls) is billed per call.

Step	What the agent calls	Reference
X posts	Keyword search with advanced operators, Latest or Top	Advanced Search
X profiles	Public profile by handle	Get User Info
Reddit	Search across public subreddits	Reddit Search
YouTube	Videos, channels, playlists	YouTube Search, YouTube SERP
Cross-platform brief	30-day ranked, clustered brief with citations	Last 30 Days
Creator discovery	Creators from a brief or a seed account	Creator AI Search, Similar Creators, KOL Creator Discovery
Reusable X workflow	Read endpoints plus OAuth-gated engagement	Twitter Autopilot
Write action (gated)	Publish a post for the authenticated user	Post Twitter

Use a Skill when the monitoring method is reusable. Call APIs directly when your application must control queries, pagination, deduplication, scoring, or storage.

Example output

Illustrative only. Handles, links, and counts are placeholders, not real posts or accounts.

## "Example App 2.0" launch: public discussion, last 14 days (EN)
Scope: X (Latest, 3 pages), Reddit (2 subreddits), YouTube (top 20 videos). Retrieved 2026-01-15.

Themes
1. Setup confusion around the new login flow: x.com/example_handle/status/000, reddit.com/r/example/comments/000
2. Praise for offline mode: youtube.com/watch?v=EXAMPLE000

Recurring questions
- "Does 2.0 still support exports?" (seen in <n> threads)

Interpretation (model, not source data)
- Early reaction skews toward usability rather than pricing.

Limits
- X results capped at 3 pages; no non-English sources; absence of discussion is not proof none exists.

Guardrails

Purpose-bound collection. Define the topic, entities, time window, platforms, and languages first. This is source-bounded analysis, not open-ended surveillance.
Preserve provenance. Normalize timestamps and metrics if needed, but keep the original URL for every record. Separate source fields, derived metrics, model labels, and missing data.
Analyze cautiously. Compare multiple records (and source types when possible) before drawing a conclusion. Missing results do not prove there is no discussion.
Privacy. Collect only what the purpose requires, respect platform terms, avoid inferring sensitive traits, and never expose credentials or unrelated personal data. Creator search can return AI-inferred gender, age range, ethnicity, and contact emails; use those fields only when lawful and necessary.
Reading is not engaging. Posting, liking, following, replying, or messaging requires OAuth and explicit approval: show the exact content and target, execute once, and verify the result. Treat automated engagement as a write action with reputational risk.
Quote, then ask. Get a free quote with batch_quote and run the same calls with batch_use only after approval. A quote checks the estimate, not the final charge. See Billing & Credits.
When AIsa may not be necessary: a direct platform integration is simpler if you use only one source, already have stable authentication and compliance controls, and do not need models or Skills.

Skills or direct APIs?

Choosing the right surface.

Sales Research Agent

Accounts, contacts, and outreach.

Authentication

Keys and OAuth.

Security & Privacy

Agent safety controls.

​The problem

​Prompt

​APIs and Skills used

​Example output

​Guardrails

​Related

Skills or direct APIs?

Sales Research Agent

Authentication

Security & Privacy

The problem

Prompt

APIs and Skills used

Example output

Guardrails

Related